Analysis And Mining Of Scientific Collaboration Behaviour Based On Scholarly Big Data

In recent academic,researching is becoming more and more complex,diverse and inter-disciplinary.Scientific collaboration has gradually become the main method for scientific re-searchers to carry out academic research and solve complex scientific problems.It is becoming the fourth age of scientific research.Collaboration behavior among scholars has received exten-sive attention from diverse fields.Scientific collaboration analysis is a typical cross-disciplinary frontier research direction,where its research fields include computational social sciences,net-work science,computer science,and science of science.Accompanied by a huge amount of academic related data,the term "scholarly big data" came into being.Scholarly big data can help people understand the academic society and scholars from the perspective of data science;promote scientific and technological progress to be more rational and efficient;help scholars dis-cover scientific research rules and mechanisms thus improving innovation ability and research efficiency;provide theoretical and methodological support for the policy makers formulating sci-entific technology development strategies,routes and guidelines.Based on the research status of scientific collaboration analysis,this dissertation aims at analyzing and exploring scientific col-laboration mechanisms and providing personalized collaboration services from the perspective of scholarly big data.The main contributions of this dissertation are as follows:1.Dynamic scientific collaboration pattern analysis.Focusing on the dynamics of scien-tific collaboration,this dissertation proposes a dynamic scientific collaboration analysis method from the perspective of academic age;analyzes how various network characteristics of ego net-works vary with scholars' academic age and studies the changing laws of scientific collaborative behaviours in different ages.The collaboration pattern distinctions between scientists form com-puter science and physics are discussed.2.Scientific collaboration sustainability prediction.In order to solve the multi-dimensional attributes of scholar entities,this dissertation proposes to profiles scholars from diver dimension-s including scholar's demographic characteristics,academic research,academic influence,and sociability;analyzes the relationships between collaboration sustainability and different dimen-sional attributes;based on the multi-dimensional scholar profiling model,predicts the collabo-ration sustainability between scholars thus improving the accuracy.3.Sustainable collaborator recommendation.For the problem that the existing scientif-ic collaborator recommendation system cannot satisfy the diversification recommendation,a sustainable collaborator recommendation method based on the conference closure is propose.Based on the scholars' common conference participation relationship and the theory of weak ties in social sciences,the conference closure mechanism in scientific collaboration is proposed and quantified.Based on the conference closure,the collaboration network is reconstructed and sustainable collaborator recommendation systems are designed.4.Lifetime collaboration Identification.Aiming at the problems of high dimension and sparseness of scientific collaboration networks,this dissertation proposes an academic collab-oration network representation learning method considering textual information.By learning research interest vector based on word embedding,the collaboration network is reconstructed.Scholar vectors can be learned based on this reconstructed networks via network embedding.While reducing the computational complexity,the similarity between scholars can be more ac-curately calculated.Based on this,a lifetime collaborator identification method is proposed.